PMS vs DMS vs AI vs a matter-intelligence layer

A practice management system is usually the operational home for matters, contacts, workflows, time, billing or accounting, depending on the product and configuration. A document management system is designed around document capture, search, versioning and access. An AI service may classify, extract, summarise or generate text, but it does not acquire professional judgement merely because its output reads fluently. A matter-intelligence layer sits across approved sources to assemble signals around a defined question and route them to a person. These categories can overlap in a modern product, so the point is not to police labels. The point is to assign authority. For every field and action, a firm should be able to say which system owns the record, which service is allowed to process it, which output is advisory and where the final decision is recorded. Without that map, convenience can become duplicated truth and unclear accountability.

An integration should not make a copied record appear more current or authoritative than its source. Map each object before connecting: matter identifiers, client records, custom fields, tasks, time entries, documents, users and permissions may all behave differently. Then decide what is read, how often it is refreshed, whether any derived value is stored and how the interface links back. Lineage should cover the source, field, observation time and transformation used to create a signal. If an AI service contributes a classification or summary, that contribution should be distinguishable from source facts. This makes correction practical. When a lawyer finds that a matter was excluded, a custom field was used inconsistently or a document was outside scope, the team can repair the responsible process instead of editing a disconnected dashboard. The design goal is a legible chain from system record to review prompt to human disposition.

A permission model is credible only when tested with real roles and representative matters. Read-only access can still expose too much if the connected account can see matters outside the Pilot cohort. A service account can become a broad bypass if the intelligence layer does not reproduce ethical walls or matter restrictions. Test positive and negative cases: a reviewer should see the matters required for the role and be unable to retrieve excluded matters, documents or client details. Record which provider personnel and subcontractors may access data, where processing occurs and which logs the firm can obtain. Australian hosting alone does not resolve these questions, and offshore processing is not automatically prohibited for every firm. Location, lawful access, contract terms, safeguards and the particular matter risk all belong in the firm's documented decision.

Procurement should include the end of the relationship. Ask what happens when API credentials are revoked, a connector is disabled, a matter changes access group or the firm leaves the service. The firm should know which data can be exported, which derived records remain, how deletion is evidenced and whether source systems continue to operate normally. It should also test rate limits, stale-data indicators, failure handling and the difference between temporary unavailability and silent omission. The Australian Signals Directorate's procurement guidance is written for particular government and large-organisation contexts, but its themes of supply-chain risk, shared responsibility, contractual protection, portability and decommissioning are useful evaluation prompts. They are not a certification shortcut. A law firm still needs a proportionate assessment of its systems, clients, legal duties and risk appetite before approving an integration.